Some years ago the Myers-Briggs personality test was popular among my peers. You can look it up, but it has introversion versus extraversion, intuitive versus sensing, thinking versus feeling as a basis for decisions, and perceiving versus judging (the latter likes to finish things up rather than hold them open). You'd get a "type" score like INTJ or ESFJ, of which there were 16 possibilities. But what I realized is that just about every human trait except for biological sex (highly bimodal) is going to be distributed on a bell curve, with most people pretty much in the middle. But "in the middle" wasn't one of the options. Myers-Briggs was useful in thinking about ways that people can be different without one being wrong. I'm pretty sure it was not validated and so thrown out like many others in favor of the Big Five model. But even on those big 5, wouldn't it be strange if they were not normally distributed too? That's not to say they're not useful because of course a significant number of people do stray significantly from the mean, and there may be interesting correlations. I just have an intuition that the answer "you're right in the middle on introversion/extraversion" or any other trait isn't what people are expecting to hear. (?).
Thanks for the comment! The little bit of research I did on personality types (or clusters) resulted in me concluding that there is very little empirical support for stable “types” which can be found in diverse populations. Like you said, our prior should probably be that scores along each dimension are more or less normally distributed, and so there’s no reason to assume identifiable types exist vs individual personalities being more or less evenly spread throughout the entire “personality space”. I think this idea is appealing anyways though because humans like to be able to slot people into clear categories
MBTI actually correlates reasonably well (.5-.7 I think) with Big 5, with E/I being Extroversion, N/S being Openness, F/T being Agreeableness, J/P being Conscientiousness, and basically ignoring Neuroticism.
It has the false-dichotomy problem you mention, but has the advantage of being well-known; nobody's going to tell you how they scored on some random internet personality test for agreeableness and conscientiousness, but lots of people know their MBTI type, and you can toss it out like a zodiac sign and gain some useful information; you won't know all their behaviors, but you have a reasonable sense of which half of the distribution they're on for four of the Big 5.
I don't know if you're already familiar with it, but trying to look at worldviews in a dimensional way may have some overlap with the "primal world beliefs" model being researched by Jer Clifton: https://myprimals.com/
Very interesting but I think it's worth pointing out that this is more just some interesting science than revealing anything useful or relevant to our lives. I don't mean that as a criticism, I love articles explaining scientific theories which this is a well written example. But I think it's worth pointing out because people tend to be interested in these personality factors for the same reason we like those 'which wizarding house do you belong to' or pseudoscientific ENTP type tests -- we're all narcissists and it feels like we are learning something profound about ourselves.
But, in reality, we are doing no such thing. We instinctively understand our friends and colleagues dispositions and likes far better than any information that could be gleaned from a personality tests. And we didn't need to do any science to know that people could vary on traits like agreeableness or extraversion and we are going to do better in most situations using our intuitions to predict how even brief acquaintances are likely to act than any model with a small number of factors will do. Of course, machine learning can be very good at predicton using very high dimensional spaces but in this case **we are the machine learning model**.
The scientific content here really just tells us about how these personality features vary across populations. That's super interesting, and no doubt useful for large institutions like the military, but facts about distribution in the population aren't all that relevant to our daily lives where we get so much information about people we interact with that for most of us it screens off information from the low dimensional factors.
I'm not suggesting you said anything else, but since it's a common misconception I felt it was worth mentioning.
--
Also, as a quibble, the number of words for various traits in a language seems kinda like a distraction. While in the analysis WaPo cites certain Inuit languages have more snow descriptive words the article suggests this isn't a particularly robust finding (depends on how you deal with compound words and synonyms etc) and even if it was the number of words in a language reflects all sorts of random factors (vocab size, number of speakers, norms about dictionaries -- french is less willing to adopt words than English).
Surely, word frequency is the more relevant measure -- and maybe that's what the other study you mentioned did, I couldn't tell -- but even that is a weird signal as it reflects both prevalence of a phenomenon, it's importance and how salient it is to other people. Are Americans more likely to be 'basic' than other cultures, do our socialization patterns means it matters more, do we just find the idea more fascinating or are just more corrupted by social media?
Yes, agreed that this isn't necessarily getting at something deep that improves prediction relative to our intuitive models about people. I think as you suggest the value is maybe there for large orgs that want to get a broad strokes understanding of personality for a large population of individuals who they cannot assess individually/don't know.
As for the number of words thing - yes, there are many reasons that we could have more words as you say... and this is likely why Cattell didn't bother to consider this when performing the factor analysis. Still, I think it's intuitive that, while word count may be too noisy of a signal to be of much use in a range of situations, I still think that more words for a thing suggests that thing is important. Even in the example he gave, that word number is clearly not a good signal of importance since we always come up with new slang for inebriation, I take a different lesson... which is that people really care about inebriation-it's a way that we bond etc. Also, I can't remember where I saw this but if you ask men and women to choose the meaning of uncommon words by multiple choice women and men know different uncommon words.... women knew all these words for different types of fabric for instance, clearly this is because we care more about this subject, men may not have invented all those words. But still, I agree with you that frequency is a better metric - but it's probably not something they could have gotten good data on at the time.
I wonder if the features identified are reflective of a fundamental organization or primarily perceptive dimensionality reduction. If you paint a tree on canvas you can realistically think about the organizing principles of a trees biology or you can Bob Ross the perceived distinguishing shapes. Why care? Something like the Lucas Critique in economics.
Dimensionality reduction of personality is predictive of situational behavior. Someone who is high in neurosis was probably more conservative about COVID avoidance, masks, gatherings etc. To what extent do differing causes of a personality trait lead to divergent optimal strategies for working with someone. One could plausibly be perceived as introverted because you are introspective or cautious of social interaction. Both causes may be classified as high introversion and behavior like quietly watching at a large gathering. The best approach to engage the two people is drastically different.
Big 5 (and dimension reduction more generally) can paint and informative picture, but do they create a useful model? I really like factor models and have used them throughout my career. Stock and Watson's factor VAR was a formative paper for me. I often think in terms of Big 5, but I am cautious of my proclivity to lean on orthogonal projections.
Yes, a lot here - one thing that you’re getting at is basically “what variance should we expect personality to explain?” - is it variance in observable behavior, internal states, what sorts of motivations drive you etc. And this question relates to your point about how personality is organized - if there are levels and some traits are more primary but the observable traits are more secondary and you could see the same trait as a result of different combinations of the more primary traits how can you pull that apart? Some of the earlier articles I read writing this were more focused on these questions of how personality is organized, which level of traits you want to try to reveal and how you could do that … but I wonder if the Big 5 has been around so long and validated so many times that people don’t bother asking these questions as much? I did see some interesting research talking about how Big 5 does or does not generalize in other populations/how approaching research using the lexical hypothesis might differ for different languages but didn’t have time to get into it too deeply - super interesting stuff though
Yes you stated the issue better than I did. My fear is that there are some important disconnects between perceived personality and building blocks of personality.
It seems like the big 5 should be mostly irrelevant here in our daily lives in predicting behavior. Our brains are already running a far more sophisticated machine learning algorithm dedicated to predicting other people that's been honed by relentless evolutionary pressure and personal experience and benefits from far more information than test questions. Looking to the big 5 rather than our intuitions is like trying to use a Bayesian text predictor to refine the output of state of the art LLMs.
In relatively unusual situations where you can't just ask people to use individual judgement -- maybe you're the army trying to assign soldiers to specialties in a uniform and unbiased fashion -- it can absolutely be useful. But in the vast majority of cases what this research is doing is groping to make explicit a low dimensional approximation of the sophisticated theory of mind we apply all the time without thinking.
Yes, I mean similarly IQ should be irrelevant to daily lives. But that doesn't mean it isn't a good metric that maps on to something real and which is predictive.
It absolutely does map to something real but the number of places where that fact is particularly useful is relatively small because it's rare that we don't have/easily get more specific information about aptitude about a task than just a generic iq score. So it's very interesting as a scientific notion at a population level but I'm skeptical about how useful it is in daily life.
1) I think personality descriptions are more about abstractions for communication.
2) I'm unconvinced people are all that great at instinctually understanding personalities different from their own or close friends. I think most folks empathize by projecting which is a bad starting place if personalities/culture/experience are quite different.
Re 1: Thats hard to evaluate. Certainly it's great to use in the scientific study of psychology but in terms of daily life it's hard to evaluate the counterfactual. I mean it might be a useful way to think about things but even without it you'd have had some model to think about things and I suspect it would be pretty good.
I mean that kinda has to be true because our ability to predict the actions of others is probably the aspect of our intelligence under the most selective pressure -- some people think it's literally why we evolved so much intelligence -- so it would be weird if a simple idea like this improved our predictive ability here alot.
Re: 2 how great do they really have to be to do better than the big 5? We may mostly project but we do modify that by the kind of broad info basic 5 give us -- if John doesn't like going out in big groups or is always turning down party invites we absolutely incorporate that information, if Mary is always very punctual and precisce we use that as well etc
The big 5 are only capturing very rough information so it doesn't take lots of accuracy to do as well just seat of the pants as you would if you were also told their big 5 scores (ofc looking at data about the specific prediction always helps).
Some years ago the Myers-Briggs personality test was popular among my peers. You can look it up, but it has introversion versus extraversion, intuitive versus sensing, thinking versus feeling as a basis for decisions, and perceiving versus judging (the latter likes to finish things up rather than hold them open). You'd get a "type" score like INTJ or ESFJ, of which there were 16 possibilities. But what I realized is that just about every human trait except for biological sex (highly bimodal) is going to be distributed on a bell curve, with most people pretty much in the middle. But "in the middle" wasn't one of the options. Myers-Briggs was useful in thinking about ways that people can be different without one being wrong. I'm pretty sure it was not validated and so thrown out like many others in favor of the Big Five model. But even on those big 5, wouldn't it be strange if they were not normally distributed too? That's not to say they're not useful because of course a significant number of people do stray significantly from the mean, and there may be interesting correlations. I just have an intuition that the answer "you're right in the middle on introversion/extraversion" or any other trait isn't what people are expecting to hear. (?).
Thanks for the comment! The little bit of research I did on personality types (or clusters) resulted in me concluding that there is very little empirical support for stable “types” which can be found in diverse populations. Like you said, our prior should probably be that scores along each dimension are more or less normally distributed, and so there’s no reason to assume identifiable types exist vs individual personalities being more or less evenly spread throughout the entire “personality space”. I think this idea is appealing anyways though because humans like to be able to slot people into clear categories
MBTI actually correlates reasonably well (.5-.7 I think) with Big 5, with E/I being Extroversion, N/S being Openness, F/T being Agreeableness, J/P being Conscientiousness, and basically ignoring Neuroticism.
It has the false-dichotomy problem you mention, but has the advantage of being well-known; nobody's going to tell you how they scored on some random internet personality test for agreeableness and conscientiousness, but lots of people know their MBTI type, and you can toss it out like a zodiac sign and gain some useful information; you won't know all their behaviors, but you have a reasonable sense of which half of the distribution they're on for four of the Big 5.
I don't know if you're already familiar with it, but trying to look at worldviews in a dimensional way may have some overlap with the "primal world beliefs" model being researched by Jer Clifton: https://myprimals.com/
Thanks Dan, no I hadn't heard of this work so will check it out!
Very interesting but I think it's worth pointing out that this is more just some interesting science than revealing anything useful or relevant to our lives. I don't mean that as a criticism, I love articles explaining scientific theories which this is a well written example. But I think it's worth pointing out because people tend to be interested in these personality factors for the same reason we like those 'which wizarding house do you belong to' or pseudoscientific ENTP type tests -- we're all narcissists and it feels like we are learning something profound about ourselves.
But, in reality, we are doing no such thing. We instinctively understand our friends and colleagues dispositions and likes far better than any information that could be gleaned from a personality tests. And we didn't need to do any science to know that people could vary on traits like agreeableness or extraversion and we are going to do better in most situations using our intuitions to predict how even brief acquaintances are likely to act than any model with a small number of factors will do. Of course, machine learning can be very good at predicton using very high dimensional spaces but in this case **we are the machine learning model**.
The scientific content here really just tells us about how these personality features vary across populations. That's super interesting, and no doubt useful for large institutions like the military, but facts about distribution in the population aren't all that relevant to our daily lives where we get so much information about people we interact with that for most of us it screens off information from the low dimensional factors.
I'm not suggesting you said anything else, but since it's a common misconception I felt it was worth mentioning.
--
Also, as a quibble, the number of words for various traits in a language seems kinda like a distraction. While in the analysis WaPo cites certain Inuit languages have more snow descriptive words the article suggests this isn't a particularly robust finding (depends on how you deal with compound words and synonyms etc) and even if it was the number of words in a language reflects all sorts of random factors (vocab size, number of speakers, norms about dictionaries -- french is less willing to adopt words than English).
Surely, word frequency is the more relevant measure -- and maybe that's what the other study you mentioned did, I couldn't tell -- but even that is a weird signal as it reflects both prevalence of a phenomenon, it's importance and how salient it is to other people. Are Americans more likely to be 'basic' than other cultures, do our socialization patterns means it matters more, do we just find the idea more fascinating or are just more corrupted by social media?
Thanks for the comment!
Yes, agreed that this isn't necessarily getting at something deep that improves prediction relative to our intuitive models about people. I think as you suggest the value is maybe there for large orgs that want to get a broad strokes understanding of personality for a large population of individuals who they cannot assess individually/don't know.
As for the number of words thing - yes, there are many reasons that we could have more words as you say... and this is likely why Cattell didn't bother to consider this when performing the factor analysis. Still, I think it's intuitive that, while word count may be too noisy of a signal to be of much use in a range of situations, I still think that more words for a thing suggests that thing is important. Even in the example he gave, that word number is clearly not a good signal of importance since we always come up with new slang for inebriation, I take a different lesson... which is that people really care about inebriation-it's a way that we bond etc. Also, I can't remember where I saw this but if you ask men and women to choose the meaning of uncommon words by multiple choice women and men know different uncommon words.... women knew all these words for different types of fabric for instance, clearly this is because we care more about this subject, men may not have invented all those words. But still, I agree with you that frequency is a better metric - but it's probably not something they could have gotten good data on at the time.
It's articles like this that make me wildly happy to be on Substack.
Thank you thank you, James!!
I wonder if the features identified are reflective of a fundamental organization or primarily perceptive dimensionality reduction. If you paint a tree on canvas you can realistically think about the organizing principles of a trees biology or you can Bob Ross the perceived distinguishing shapes. Why care? Something like the Lucas Critique in economics.
Dimensionality reduction of personality is predictive of situational behavior. Someone who is high in neurosis was probably more conservative about COVID avoidance, masks, gatherings etc. To what extent do differing causes of a personality trait lead to divergent optimal strategies for working with someone. One could plausibly be perceived as introverted because you are introspective or cautious of social interaction. Both causes may be classified as high introversion and behavior like quietly watching at a large gathering. The best approach to engage the two people is drastically different.
Big 5 (and dimension reduction more generally) can paint and informative picture, but do they create a useful model? I really like factor models and have used them throughout my career. Stock and Watson's factor VAR was a formative paper for me. I often think in terms of Big 5, but I am cautious of my proclivity to lean on orthogonal projections.
Yes, a lot here - one thing that you’re getting at is basically “what variance should we expect personality to explain?” - is it variance in observable behavior, internal states, what sorts of motivations drive you etc. And this question relates to your point about how personality is organized - if there are levels and some traits are more primary but the observable traits are more secondary and you could see the same trait as a result of different combinations of the more primary traits how can you pull that apart? Some of the earlier articles I read writing this were more focused on these questions of how personality is organized, which level of traits you want to try to reveal and how you could do that … but I wonder if the Big 5 has been around so long and validated so many times that people don’t bother asking these questions as much? I did see some interesting research talking about how Big 5 does or does not generalize in other populations/how approaching research using the lexical hypothesis might differ for different languages but didn’t have time to get into it too deeply - super interesting stuff though
Yes you stated the issue better than I did. My fear is that there are some important disconnects between perceived personality and building blocks of personality.
It seems like the big 5 should be mostly irrelevant here in our daily lives in predicting behavior. Our brains are already running a far more sophisticated machine learning algorithm dedicated to predicting other people that's been honed by relentless evolutionary pressure and personal experience and benefits from far more information than test questions. Looking to the big 5 rather than our intuitions is like trying to use a Bayesian text predictor to refine the output of state of the art LLMs.
In relatively unusual situations where you can't just ask people to use individual judgement -- maybe you're the army trying to assign soldiers to specialties in a uniform and unbiased fashion -- it can absolutely be useful. But in the vast majority of cases what this research is doing is groping to make explicit a low dimensional approximation of the sophisticated theory of mind we apply all the time without thinking.
Yes, I mean similarly IQ should be irrelevant to daily lives. But that doesn't mean it isn't a good metric that maps on to something real and which is predictive.
It absolutely does map to something real but the number of places where that fact is particularly useful is relatively small because it's rare that we don't have/easily get more specific information about aptitude about a task than just a generic iq score. So it's very interesting as a scientific notion at a population level but I'm skeptical about how useful it is in daily life.
Two divergent thoughts:
1) I think personality descriptions are more about abstractions for communication.
2) I'm unconvinced people are all that great at instinctually understanding personalities different from their own or close friends. I think most folks empathize by projecting which is a bad starting place if personalities/culture/experience are quite different.
Re 1: Thats hard to evaluate. Certainly it's great to use in the scientific study of psychology but in terms of daily life it's hard to evaluate the counterfactual. I mean it might be a useful way to think about things but even without it you'd have had some model to think about things and I suspect it would be pretty good.
I mean that kinda has to be true because our ability to predict the actions of others is probably the aspect of our intelligence under the most selective pressure -- some people think it's literally why we evolved so much intelligence -- so it would be weird if a simple idea like this improved our predictive ability here alot.
Re: 2 how great do they really have to be to do better than the big 5? We may mostly project but we do modify that by the kind of broad info basic 5 give us -- if John doesn't like going out in big groups or is always turning down party invites we absolutely incorporate that information, if Mary is always very punctual and precisce we use that as well etc
The big 5 are only capturing very rough information so it doesn't take lots of accuracy to do as well just seat of the pants as you would if you were also told their big 5 scores (ofc looking at data about the specific prediction always helps).
Cool picture.
ChatGPT can do great things ;)